Tool/software: TI-RTOS
Our model target is doing GoogleNet + Yolo V3, so we need TI to help us clarify the following questions.
Current SDK v3.03 is using SSD model, is there any possible to replace SSD in TIDL module, for example, Yolo V3 (You Only Look Once: Unified, Real-Time Object Detection, CVPR 2016)
If yes, we can replace SSD by our own model
Fig.1 is TI TDA2 architecture what we supposed and the yellow block is our customized model.
Q1: if TI architecture is not like above diagram, can you provide your architecture for our reference?
Fig.1
Q2: How can we implement our model into TI TDA2?
For TIDL input and output, can you point me two issues solution.
l We need to know Input BGR input file format and seg_out.bin output file format
l We need API function such as CUDA level in TI platform for hardware accelerate.
2.If not, we can NOT replace SSD by our own model and both issues of Q2 cannot be resolved
Q3: when does TI platform will release Yolo algorithm? If not, how can we build our Yolo library module in TI platform?
Q4: how can we apply other (Squeezent、Googlenet、Mobilent) in TI platform, any other example. If possible, can we apply yolo+(google net or VGG) in TI platform?
